apparmor/parser/parser_regex.c

/* $Id$ */

/*
 *   Copyright (c) 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007
 *   NOVELL (All rights reserved)
 *
 *   This program is free software; you can redistribute it and/or
 *   modify it under the terms of version 2 of the GNU General Public
 *   License published by the Free Software Foundation.
 *
 *   This program is distributed in the hope that it will be useful,
 *   but WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *   GNU General Public License for more details.
 *
 *   You should have received a copy of the GNU General Public License
 *   along with this program; if not, contact Novell, Inc.
 */

#include <stdio.h>
#include <stdarg.h>
#include <string.h>
#include <libintl.h>
#define _(s) gettext(s)

/* #define DEBUG */

#include "parser.h"
#include "libapparmor_re/apparmor_re.h"

enum error_type {
	e_no_error,
	e_parse_error,
	e_buffer_overflow
};

/* filter_escapes: scan input for any escape characters
 * and remove, and reduce double \\ to a single
 * NOTE: modifies in place the contents of the name argument */
static void filter_escapes(char *name)
{
	char *sptr, *dptr;
	BOOL bEscape = 0;

	if (!name)	/* shouldn't happen */
		return;

	sptr = dptr = name;
	while (*sptr) {
		if (*sptr == '\\') {
			if (bEscape) {
				*dptr++ = *sptr++;
			} else {
				++sptr;
				bEscape = TRUE;
				continue;
			}
		} else if (dptr < sptr) {
			*dptr++ = *sptr++;
		} else {
			dptr++;
			sptr++;
		}
		bEscape = 0;
	}
	*dptr = 0;
}

/* Filters out multiple slashes (except if the first two are slashes,
 * that's a distinct namespace in linux) and trailing slashes.
 * NOTE: modifies in place the contents of the path argument */

static void filter_slashes(char *path)
{
	char *sptr, *dptr;
	BOOL seen_slash = 0;
	int len;

	if (!path || (strlen(path) < 2))
		return;

	sptr = dptr = path;

	/* special case for linux // namespace */
	if (sptr[0] == '/' && sptr[1] == '/' && sptr[2] != '/') {
		sptr += 2;
		dptr += 2;
	}

	while (*sptr) {
		if (*sptr == '/') {
			if (seen_slash) {
				++sptr;
			} else {
				*dptr++ = *sptr++;
				seen_slash = TRUE;
			}
		} else {
			seen_slash = 0;
			if (dptr < sptr) {
				*dptr++ = *sptr++;
			} else {
				dptr++;
				sptr++;
			}
		}
	}
	*dptr = 0;

	if (regex_type != AARE_DFA) {
		/* eliminate trailing slashes for versions of apparmor that
		 * do not use the dfa engine.
		 * Versions of apparmor which use the dfa engine use the
		 * trailing / to differentiate between file and directory
		 * matches
		 */
		len = strlen(path);
		if (len > 2 && path[len -1] == '/') {
			path[len - 1] = 0;
		}
	}
}

static pattern_t convert_aaregex_to_pcre(const char *aare, int anchor,
					 char *pcre, size_t pcre_size,
					 int *first_re_pos)
{
#define STORE(_src, _dest, _len) \
	if ((const char*)_dest + _len > (pcre + pcre_size)){ \
		error = e_buffer_overflow; \
	} else { \
		memcpy(_dest, _src, _len); \
		_dest += _len; \
	}
#define update_re_pos(X) if (!(*first_re_pos)) { *first_re_pos = (X); }

	*first_re_pos = 0;

	int ret = TRUE;
	/* flag to indicate input error */
	enum error_type error;

	const char *sptr;
	char *dptr;
	pattern_t ptype;

	BOOL bEscape = 0;	/* flag to indicate escape */
	int ingrouping = 0;	/* flag to indicate {} context */
	int incharclass = 0;	/* flag to indicate [ ] context */

	error = e_no_error;
	ptype = ePatternBasic;	/* assume no regex */

	sptr = aare;
	dptr = pcre;

	if (dfaflags & DFA_DUMP_RULE_EXPR)
		fprintf(stderr, "aare: %s   ->   ", aare);

	if (anchor)
		/* anchor beginning of regular expression */
		*dptr++ = '^';

	while (error == e_no_error && *sptr) {
		switch (*sptr) {

		case '\\':
			/* concurrent escapes are allowed now and
			 * output as two consecutive escapes so that
			 * pcre won't interpret them
			 * \\\{...\\\} will be emitted as \\\{...\\\}
			 * for pcre matching.  For string matching
			 * and globbing only one escape is output
			 * this is done by stripping later
			 */
			if (bEscape) {
				STORE("\\\\", dptr, 2);
			} else {
				bEscape = TRUE;
				++sptr;
				continue;	/*skip turning bEscape off */
			}	/* bEscape */
			break;
		case '*':
			if (bEscape) {
				/* '*' is a PCRE special character */
				/* We store an escaped *, in case we
				 * end up using this regex buffer (i.e another
				 * non-escaped regex follows)
				 */
				STORE("\\*", dptr, 2);
			} else {
				if ((dptr > pcre) &&  *(dptr - 1) == '/') {
					#if 0
					// handle comment containing use
					// of C comment characters
					// /* /*/ and /** to describe paths
					//
					// modify what is emitted for * and **
					// when used as the only path
					// component
					// ex.
					// /* /*/ /**/ /**
					// this prevents these expressions
					// from matching directories or
					// invalid paths
					// in these case * and ** must
					// match at least 1 character to
					// get a valid path element.
					// ex.
					// /foo/* -> should not match /foo/
					// /foo/*bar -> should match /foo/bar
					// /*/foo -> should not match //foo
					#endif
					const char *s = sptr;
					while (*s == '*')
						s++;
					if (*s == '/' || !*s) {
						STORE("[^/\\x00]", dptr, 8);
					}
				}
				if (*(sptr + 1) == '*') {
					/* is this the first regex form we
					 * have seen and also the end of
					 * pattern? If so, we can use
					 * optimised tail globbing rather
					 * than full regex.
					 */
					update_re_pos(sptr - aare);
					if (*(sptr + 2) == '\0' &&
					    ptype == ePatternBasic) {
						ptype = ePatternTailGlob;
					} else {
						ptype = ePatternRegex;
					}

					STORE("[^\\x00]*", dptr, 8);
					sptr++;
				} else {
					update_re_pos(sptr - aare);
					ptype = ePatternRegex;
					STORE("[^/\\x00]*", dptr, 9);
				}	/* *(sptr+1) == '*' */
			}	/* bEscape */

			break;

		case '?':
			if (bEscape) {
				/* ? is not a PCRE regex character
				 * so no need to escape, just skip
				 * transform
				 */
				STORE(sptr, dptr, 1);
			} else {
				update_re_pos(sptr - aare);
				ptype = ePatternRegex;
				STORE("[^/\\x00]", dptr, 8);
			}
			break;

		case '[':
			if (bEscape) {
				/* [ is a PCRE special character */
				STORE("\\[", dptr, 2);
			} else {
				update_re_pos(sptr - aare);
				incharclass = 1;
				ptype = ePatternRegex;
				STORE(sptr, dptr, 1);
			}
			break;

		case ']':
			if (bEscape) {
				/* ] is a PCRE special character */
				STORE("\\]", dptr, 2);
			} else {
				incharclass = 0;
				STORE(sptr, dptr, 1);
			}
			break;

		case '{':
			if (bEscape) {
				/* { is a PCRE special character */
				STORE("\\{", dptr, 2);
			} else {
				if (ingrouping) {
					error = e_parse_error;
					PERROR(_("%s: Illegal open {, nesting groupings not allowed\n"),
					       progname);
				} else {
					update_re_pos(sptr - aare);
					ingrouping = 1;
					ptype = ePatternRegex;
					STORE("(", dptr, 1);
				}
			}
			break;

		case '}':
			if (bEscape) {
				/* { is a PCRE special character */
				STORE("\\}", dptr, 2);
			} else {
				if (ingrouping <= 1) {

					error = e_parse_error;

					if (ingrouping == 1) {
						PERROR(_("%s: Regex grouping error: Invalid number of items between {}\n"),
						       progname);

						ingrouping = 0;	/* prevent further errors */

					} else {	/* ingrouping == 0 */
						PERROR(_("%s: Regex grouping error: Invalid close }, no matching open { detected\n"),
						       progname);
					}
				} else {	/* ingrouping > 1 */
					ingrouping = 0;
					STORE(")", dptr, 1);
				}
			}	/* bEscape */

			break;

		case ',':
			if (bEscape) {
				/* , is not a PCRE regex character
				 * so no need to escape, just skip
				 * transform
				 */
				STORE(sptr, dptr, 1);
			} else {
				if (ingrouping) {
					++ingrouping;
					STORE("|", dptr, 1);
				} else {
					STORE(sptr, dptr, 1);
				}
			}
			break;

			/* these are special outside of character
			 * classes but not in them */
		case '^':
		case '$':
			if (incharclass) {
				STORE(sptr, dptr, 1);
			} else {
				STORE("\\", dptr, 1);
				STORE(sptr, dptr, 1);
			}
			break;

			/*
			 * Not a subdomain regex, but needs to be
			 * escaped as it is a pcre metacharacter which
			 * we don't want to support. We always escape
			 * these, so no need to check bEscape
			 */
		case '.':
		case '+':
		case '|':
		case '(':
		case ')':
			STORE("\\", dptr, 1);
			// fall through to default

		default:
			STORE(sptr, dptr, 1);
			break;
		}	/* switch (*sptr) */

		bEscape = FALSE;
		++sptr;
	}		/* while error == e_no_error && *sptr) */

	if (ingrouping > 0 || incharclass) {
		error = e_parse_error;

		PERROR(_("%s: Regex grouping error: Unclosed grouping or character class, expecting close }\n"),
		       progname);
	}

	/* anchor end and terminate pattern string */
	if ((error == e_no_error) && anchor) {
		STORE("$" , dptr, 1);
	}
	if (error == e_no_error) {
		STORE("", dptr, 1);
	}
	/* check error  again, as above STORE may have set it */
	if (error != e_no_error) {
		if (error == e_buffer_overflow) {
			PERROR(_("%s: Internal buffer overflow detected, %d characters exceeded\n"),
			       progname, PATH_MAX);
		}

		PERROR(_("%s: Unable to parse input line '%s'\n"),
		       progname, aare);

		ret = FALSE;
		goto out;
	}

out:
	if (ret == FALSE)
		ptype = ePatternInvalid;

	if (dfaflags & DFA_DUMP_RULE_EXPR)
		fprintf(stderr, "%s\n", pcre);

	return ptype;
}

static int process_pcre_entry(struct cod_entry *entry)
{
	char tbuf[PATH_MAX + 3];	/* +3 for ^, $ and \0 */
	int ret = TRUE;
	pattern_t ptype;
	int pos;
	if (!entry) 		/* shouldn't happen */
		return TRUE;

	ptype = convert_aaregex_to_pcre(entry->name, 1, tbuf, PATH_MAX+3, &pos);
	if (ptype == ePatternInvalid)
		return FALSE;

	entry->pattern_type = ptype;

	/*
	 * Only use buffer (tbuf) that we built above, if we
	 * identified a pattern requiring full regex support.
	 */
	if (ptype == ePatternRegex) {
		int pattlen = strlen(tbuf);

		if ((entry->pat.regex = malloc(pattlen + 1))) {
			const char *errorreason;
			int errpos;

			strcpy(entry->pat.regex, tbuf);

			if ((entry->pat.compiled =
			     pcre_compile(entry->pat.regex, 0,
					  &errorreason, &errpos,
					  NULL))) {
				/* NULL out tables, kernel will use a
				 * private version
				 */
				entry->pat.compiled->tables = NULL;
			} else {
				int i;

				PERROR(_("%s: Failed to compile regex '%s' [original: '%s']\n"),
				       progname, entry->pat.regex,
				       entry->name);

				PERROR(_("%s: error near               "),
				       progname);

				for (i = 0; i < errpos; i++) {
					fputc('.', stderr);
				}

				fputc('^', stderr);
				fputc('\n', stderr);

				PERROR(_("%s: error reason: '%s'\n"),
				       progname, errorreason);

				free(entry->pat.regex);
				entry->pat.regex = NULL;

				ret = FALSE;
			}
		} else {
			PERROR(_("%s: Failed to compile regex '%s' [original: '%s'] - malloc failed\n"),
			       progname, entry->pat.regex, entry->name);

			ret = FALSE;
		}
	} else {
		/* not a regex, scan input for any escape characters
		 * and remove, and reduce double \\ to a single */
		filter_escapes(entry->name);
	}		/* ptype == ePatternRegex */

	return ret;
}

static const char *local_name(const char *name)
{
	const char *t;

	for (t = strstr(name, "//") ; t ; t = strstr(name, "//"))
		name = t + 2;

	return name;
}

static int process_profile_name_xmatch(struct codomain *cod)
{
	char tbuf[PATH_MAX + 3];	/* +3 for ^, $ and \0 */
	pattern_t ptype;
	const char *name;

	/* don't filter_slashes for profile names */
	name = local_name(cod->name);
	ptype = convert_aaregex_to_pcre(name, 0, tbuf, PATH_MAX + 3,
					&cod->xmatch_len);

	if (ptype == ePatternInvalid) {
		PERROR(_("%s: Invalid profile name '%s' - bad regular expression\n"), progname, name);
		return FALSE;
	} else if (ptype == ePatternBasic && !cod->altnames) {
		/* no regex so do not set xmatch */
		cod->xmatch = NULL;
		cod->xmatch_len = 0;
		cod->xmatch_size = 0;
	} else {
		/* build a dfa */
		aare_ruleset_t *rule = aare_new_ruleset(0);
		if (!rule)
			return FALSE;
		if (!aare_add_rule(rule, tbuf, 0, AA_MAY_EXEC, 0, dfaflags)) {
			aare_delete_ruleset(rule);
			return FALSE;
		}
		if (cod->altnames) {
			struct alt_name *alt;
			list_for_each(cod->altnames, alt) {
				int len;
				convert_aaregex_to_pcre(alt->name, 0, tbuf,
							PATH_MAX + 3, &len);
				if (!aare_add_rule(rule, tbuf, 0, AA_MAY_EXEC, 0, dfaflags)) {
					aare_delete_ruleset(rule);
					return FALSE;
				}
			}
		}
		cod->xmatch = aare_create_dfa(rule, &cod->xmatch_size,
					      dfaflags);
		aare_delete_ruleset(rule);
		if (!cod->xmatch)
			return FALSE;
	}

	return TRUE;
}

static int process_dfa_entry(aare_ruleset_t *dfarules, struct cod_entry *entry)
{
	char tbuf[PATH_MAX + 3];	/* +3 for ^, $ and \0 */
	pattern_t ptype;
	int pos;

	if (!entry) 		/* shouldn't happen */
		return TRUE;


	if (entry->mode & ~AA_CHANGE_PROFILE)
		filter_slashes(entry->name);
	ptype = convert_aaregex_to_pcre(entry->name, 0, tbuf, PATH_MAX+3, &pos);
	if (ptype == ePatternInvalid)
		return FALSE;

	entry->pattern_type = ptype;

	/* ix implies m but the apparmor module does not add m bit to
	 * dfa states like it does for pcre
	 */
	if ((entry->mode >> AA_OTHER_SHIFT) & AA_EXEC_INHERIT)
		entry->mode |= AA_EXEC_MMAP << AA_OTHER_SHIFT;
	if ((entry->mode >> AA_USER_SHIFT) & AA_EXEC_INHERIT)
		entry->mode |= AA_EXEC_MMAP << AA_USER_SHIFT;

	/* relying on ptrace and change_profile not getting merged earlier */

	/* the link bit on the first pair entry should not get masked
	 * out by a deny rule, as both pieces of the link pair must
	 * match.  audit info for the link is carried on the second
	 * entry of the pair
	 */
	if (entry->deny && (entry->mode & AA_LINK_BITS)) {
		if (!aare_add_rule(dfarules, tbuf, entry->deny,
				   entry->mode & ~AA_LINK_BITS,
				   entry->audit & ~AA_LINK_BITS, dfaflags))
			return FALSE;
	} else if (entry->mode & ~AA_CHANGE_PROFILE) {
		if (!aare_add_rule(dfarules, tbuf, entry->deny, entry->mode,
				   entry->audit, dfaflags))
			return FALSE;
	}

	if (entry->mode & (AA_LINK_BITS)) {
		/* add the pair rule */
		char lbuf[PATH_MAX + 8];
		int perms = AA_LINK_BITS & entry->mode;
		char *vec[2];
		int pos;
		vec[0] = tbuf;
		if (entry->link_name) {
			ptype = convert_aaregex_to_pcre(entry->link_name, 0, lbuf, PATH_MAX + 8, &pos);
			if (ptype == ePatternInvalid)
				return FALSE;
			if (entry->subset)
				perms |= LINK_TO_LINK_SUBSET(perms);
			vec[1] = lbuf;
		} else {
			perms |= LINK_TO_LINK_SUBSET(perms);
			vec[1] = "/[^/].*";
		}
		if (!aare_add_rule_vec(dfarules, entry->deny, perms, entry->audit & AA_LINK_BITS, 2, vec, dfaflags))
			return FALSE;
	}
	if (entry->mode & AA_CHANGE_PROFILE) {
		if (entry->namespace) {
			char *vec[2];
			char lbuf[PATH_MAX + 8];
			int pos;
			ptype = convert_aaregex_to_pcre(entry->namespace, 0, lbuf, PATH_MAX + 8, &pos);
			vec[0] = lbuf;
			vec[1] = tbuf;
			if (!aare_add_rule_vec(dfarules, 0, AA_CHANGE_PROFILE, 0, 2, vec, dfaflags))
			    return FALSE;
		} else {
		  if (!aare_add_rule(dfarules, tbuf, 0, AA_CHANGE_PROFILE, 0, dfaflags))
				return FALSE;
		}
	}
	if (entry->mode & (AA_USER_PTRACE | AA_OTHER_PTRACE)) {
		int mode = entry->mode & (AA_USER_PTRACE | AA_OTHER_PTRACE);
		if (entry->namespace) {
			char *vec[2];
			vec[0] = entry->namespace;
			vec[1] = entry->name;
			if (!aare_add_rule_vec(dfarules, 0, mode, 0, 2, vec, dfaflags))
			    return FALSE;
		} else {
		  if (!aare_add_rule(dfarules, entry->name, 0, mode, 0, dfaflags))
				return FALSE;
		}
	}
	return TRUE;
}

int post_process_entries(struct codomain *cod)
{
	int ret = TRUE, rc;
	struct cod_entry *entry;
	int count = 0;

	list_for_each(cod->entries, entry) {
		if (regex_type == AARE_DFA) {
			rc = process_dfa_entry(cod->dfarules, entry);
		} else {
			filter_slashes(entry->name);
			rc = process_pcre_entry(entry);
		}
		if (!rc)
			ret = FALSE;
		count++;
	}

	cod->dfarule_count = count;
	return ret;
}

int process_regex(struct codomain *cod)
{
	int error = -1;

	if (regex_type == AARE_DFA) {
		if (!process_profile_name_xmatch(cod))
			goto out;

		cod->dfarules = aare_new_ruleset(0);
		if (!cod->dfarules)
			goto out;
	}
	if (!post_process_entries(cod))
		goto out;

	if (regex_type == AARE_DFA && cod->dfarule_count > 0) {
		cod->dfa = aare_create_dfa(cod->dfarules, &cod->dfa_size,
					   dfaflags);
		aare_delete_ruleset(cod->dfarules);
		cod->dfarules = NULL;
 		if (!cod->dfa)
			goto out;
/*
		if (cod->dfa_size == 0) {
			PERROR(_("profile %s: has merged rules (%s) with "
				 "multiple x modifiers\n"),
			       cod->name, (char *) cod->dfa);
			free(cod->dfa);
			cod->dfa = NULL;
			goto out;
		}
*/
	}
	/*
	 * Post process subdomain(s):
	 *
	 * They are chained from the toplevel subdomain pointer
	 * thru each <codomain> next pointer.

	 * i.e first subdomain is list->subdomain
	 *    second subdomain is list->subdomain->next
	 *
	 * N.B sub-subdomains are not valid so:
	 * if (list->subdomain) {
	 *    assert(list->subdomain->subdomain == NULL)
	 * }
	 */
	if (process_hat_regex(cod) != 0)
		goto out;

	error = 0;

out:
	return error;
}

void reset_regex(void)
{
	aare_reset_matchflags();
}

#ifdef UNIT_TEST
#define MY_TEST(statement, error)		\
	if (!(statement)) {			\
		PERROR("FAIL: %s\n", error);	\
		rc = 1;				\
	}

/* Guh, fake routine */
void yyerror(char *msg, ...)
{
	va_list arg;
	char buf[PATH_MAX];

	va_start(arg, msg);
	vsnprintf(buf, sizeof(buf), msg, arg);
	va_end(arg);

	PERROR(_("AppArmor parser error: %s\n"), buf);

	exit(1);
}
/* Guh, fake symbol */
char *progname;

static int test_filter_escapes(void)
{
	int rc = 0;
	char *test_string;

	test_string = strdup("foo\\\\foo");
	filter_escapes(test_string);
	MY_TEST(strcmp(test_string, "foo\\foo") == 0, "simple filter for \\\\");

	test_string = strdup("foo\\foo");
	filter_escapes(test_string);
	MY_TEST(strcmp(test_string, "foofoo") == 0, "simple filter for \\f");
	return rc;
}

static int test_filter_slashes(void)
{
	int rc = 0;
	char *test_string;

	test_string = strdup("///foo//////f//oo////////////////");
	filter_slashes(test_string);
	MY_TEST(strcmp(test_string, "/foo/f/oo") == 0, "simple tests");

	test_string = strdup("/foo/f/oo");
	filter_slashes(test_string);
	MY_TEST(strcmp(test_string, "/foo/f/oo") == 0, "simple test for no changes");

	test_string = strdup("/");
	filter_slashes(test_string);
	MY_TEST(strcmp(test_string, "/") == 0, "simple test for '/'");

	test_string = strdup("");
	filter_slashes(test_string);
	MY_TEST(strcmp(test_string, "") == 0, "simple test for ''");

	test_string = strdup("//usr");
	filter_slashes(test_string);
	MY_TEST(strcmp(test_string, "//usr") == 0, "simple test for // namespace");

	test_string = strdup("//");
	filter_slashes(test_string);
	MY_TEST(strcmp(test_string, "//") == 0, "simple test 2 for // namespace");

	test_string = strdup("///usr");
	filter_slashes(test_string);
	MY_TEST(strcmp(test_string, "/usr") == 0, "simple test for ///usr");

	test_string = strdup("///");
	filter_slashes(test_string);
	MY_TEST(strcmp(test_string, "/") == 0, "simple test for ///");

	test_string = strdup("/a/");
	filter_slashes(test_string);
	MY_TEST(strcmp(test_string, "/a") == 0, "simple test for /a/");

	return rc;
}

int regex_type = AARE_PCRE;

int main(void)
{
	int rc = 0;
	int retval;

	retval = test_filter_escapes();
	if (retval != 0)
		rc = retval;

	retval = test_filter_slashes();
	if (retval != 0)
		rc = retval;

	return rc;
}
#endif /* UNIT_TEST */